The LIFE study was a prospective, randomized, double-masked, parallel group study (n = 9,193) that evaluated the long-term effects of losartan- compared with atenolol-based antihypertensive therapy on cardiovascular morbidity and mortality in patients with hypertension and electrocardiographic (ECG)-LVH. The LIFE study design, organization, clinical measures, endpoint definitions, basis for choice of comparative agents, statistical power calculations, recruitment details, baseline characteristics, 1-year follow-up, and primary results have been published [31–32]. The LIFE patients with AF at study baseline (n = 342) had a lower rate of cardiovascular events with losartan- compared with atenolol-based treatment [1]. The present study was done in 8,243 LIFE participants with ECG-documented sinus rhythm, no history of AF, and available SUA measurements at study baseline.

As previously described, patients aged 55–80 years with previously treated or untreated hypertension and ECG signs of LVH were randomized to initial therapy with 50 mg/day of losartan or atenolol after 1–2 weeks on placebo if they had sitting systolic BP 160–200 mmHg and/or diastolic BP of 95–115 mmHg [32]. In both groups, hydrochlorothiazide 12.5 mg was added in case of insufficient BP lowering, after which study drug was increased to 100 mg/day and supplemented with hydrochlorothiazide 25 mg or additional antihypertensive therapy to reach target BP of < 140/90 mmHg. Patients were enrolled from June 1995 to May 1997 and were followed ≥ 4 years (mean 5.0 ± 0.4 years) with regular visits. Sitting BP was recorded 24 h post-dose (range 22–26 h) and blood was obtained for SUA determination at baseline and at annual revisits during the study.

New-onset AF was documented by annual in-study ECGs by Minnesota codes including all types of AF [33] and determination of ECG-LVH by both Cornell voltage-duration product and Sokolow-Lyon voltage at a central ECG core laboratory as previously described [31, 32]. Supplemental analyses used incident AF reported by investigators during the follow-up period, either alone or in combination with new AF by Minnesota code (n = 652). Care of patients’ new onset AF was left to the discretion of local investigators. Laboratory measurements of SUA were made at central laboratories for the Nordic countries (located in Brussels, Belgium) and in the United States that assured comparability of measurements by cross-validation.

The authors had full access to the data and take responsibility for its integrity. All authors have read and agreed to the manuscript as written. SPSS version 12.0 (SPSS, Inc., Chicago, IL, USA) was used for statistical analysis. Potential risk factors (including baseline clinical, demographic, and laboratory data) were assessed for association with new-onset AF. Cox proportional hazards models were used to compare hazard ratios (HRs) between study treatment allocation groups (losartan or atenolol) and to evaluate contributions of differences in the degree of LVH (both Cornell voltage-duration product and Sokolow-Lyon voltage as continuous variables), Framingham risk score, and other covariates. For each baseline characteristic, a univariate proportional hazards regression model was used to estimate the HR and its 95% confidence interval (CI). Variables without significant associations with new-onset AF were eliminated before developing multivariable models.

Multivariable analyses were performed using Cox regression models with inclusion of remaining baseline variables to identify those independently associated with the endpoints. Two-tailed P < 0.05 was considered significant. Multivariable Cox models were constructed using SUA and other variables re-measured at intervals during LIFE follow-up as continuous time-varying covariates to take into account variable changes of SUA from baseline among patients related to losartan treatment, changes in renal function, and aging. Supplemental Cox models using quartile ranges of SUA measurements at baseline were used to assess the predictive value of SUA levels in defined ranges for subsequent new AF during intensive antihypertensive treatment in LIFE. HRs were calculated to assess the risk of developing new-onset AF in patients whose SUA fell in the 1st through 4th quartiles at baseline using sex-specific 2nd, 3rd, and 4th SUA quartile ranges as follows: < 309, 309–353, 354–408, and ≥ 409 μmol/L for men; and < 255, 255–297, 298–345, and ≥ 346 μmol/L for women. The 1st through the 4th quartiles of baseline SUA contained n = 2,033, 2,070, 2,100, and 2,040 patients, respectively, at enrolment. These analyses were repeated using SUA within these sex-specific ranges at each visit during the LIFE study as a time-varying predictor of subsequent AF. A modified Kaplan-Meier plot [34] was constructed to illustrate the association between time-varying SUA in these ranges and the rate of new-onset AF.

The present study was performed in 8,243 LIFE participants with ECG-documented sinus rhythm, no history of AF, and available SUA measurements at study baseline. The eligible subjects had a mean age of 66.8 years and 46% were women. Patients treated with losartan had, compared with those on atenolol, similar baseline characteristics with respect to proportion of male sex (54% vs. 55%, respectively), mean age (66.8 years in each group), mean Framingham risk score (22.1 vs. 22.3, respectively), history of diabetes (13% in each group), and mean baseline systolic BP (174 mmHg in each group).

Assessment of univariate relations between baseline variables and new-onset AF (Table 1) identified significant positive associations with older age, non-black ethnicity, higher systolic BP, presence of LVH by Cornell voltage-duration product criteria, prevalent coronary heart disease and cardiovascular disease, lower serum potassium, and higher urinary creatinine/albumin ratio. New-onset AF was not associated with baseline smoking habit, total cholesterol, high-density-lipoprotein (HDL)-cholesterol, or plasma glucose levels. Similarly, baseline SUA level did not predict subsequent incident AF (P > 0.20).

Because SUA and BP levels both changed substantially during the LIFE study, a further multivariable Cox model considered SUA and systolic BP as time-varying covariates. In this model, time-varying SUA strongly predicted subsequent new-onset AF by Minnesota code [HR = 1.19 per 16.8 μmol/L (1 mg/dL), (95% CIs, 1.12–1.26), P < 0.0001] independent of the positive associations of new AF with higher Cornell voltage-duration product [HR = 1.10 per 1,023 ms·mm (95% CIs, 1.01–1.21), P = 0.034], older age [HR = 1.95 per 7.0 years (95% CIs, 1.73–2.20), P < 0.0001], male sex [HR = 1.46 (95% CIs, 1.09–1.94), P = 0.010], and the protective effect of losartan treatment [HR = 0.75 (95% CIs, 0.61–0.93), P = 0.007]. Systolic BP, Framingham risk score, and previous cardiovascular disease did not contribute significantly to this model.

SUA at baseline and at each annual follow-up visit was divided into four groups based on sex-specific quartile ranges of SUA at baseline, as described above. Additional univariate and multivariable Cox models that adjusted for the same covariates were used to compare SUA level in quartiles with new-onset AF. Patients were analyzed as belonging to the quartiles they belonged to from one year to the next thus the term “time-varying”. SUA in quartile 4 was strongly predictive of new-onset AF defined by Minnesota code in the univariate as well as in multivariate analyses (Table 2).

In a Cox model that considered sex-specific SUA level partitions and systolic BP as time-varying covariates and included time-varying Cornell voltage-duration product, Framingham risk score, and age as well as categorical variables of baseline cerebrovascular disease, gender, and treatment group, SUA in the 4th quartile strongly predicted new-onset AF by Minnesota Code [HR = 1.65 (95% CIs, 1.19–2.29), P = 0.003] independent of effects of systolic BP [HR = 0.92 per 14 mmHg (95% CIs, 0.84–1.00), P = 0.044], Cornell voltage-duration product [HR = 1.10 per 1,023 ms·mm (95% CIs, 1.01–1.21), P = 0.033], age [HR = 1.94 per 7.0 years (95% CIs, 1.72–2.19), P ≤ 0.0001], male gender [HR = 1.44 (95% CIs, 1.08–1.92), P = 0.014], and atenolol treatment [HR = 1.35 (95% CIs, 1.10–1.66), P = 0.005]. Framingham risk score and cerebrovascular disease, which were significant in the univariate analysis, did not contribute significantly to this multivariable analysis (Table 3). Of note, SUA was second only to age as a strong predictor of new-onset AF, as shown by the Wald statistic of 30, a four-fold greater effect than that of the next most potent predictor.

Additional time-varying covariate analyses comparing the 2nd, 3rd, and 4th sex specific quartiles of SUA with the 1st quartile separately as predictors of new-onset AF by Minnesota code, with adjustment for time-varying systolic BP, age, sex, prevalent cerebrovascular disease, Cornell voltage-duration product, and Framingham risk score at baseline in losartan-treated and atenolol-treated patients were performed. The HRs in losartan-treated patients [HR = 0.82 (95% CIs, 0.43–1.47), P = 0.50, HR = 1.12 (95% CIs, 0.65–1.92), P = 0.68, and HR = 1.88 (95% CIs, 1.17–3.01), P = 0.009] and in atenolol-treated patients [HR = 0.99 (95% CIs, 0.55–1.76), P = 0.96, HR = 1.18 (95% CIs, 0.69–2.20), P = 0.55, and HR = 1.59 (95% CIs, 0.92–2.58), P = 0.065] did not suggest a major inter-treatment difference in the association of higher SUA with AF. The use of diuretic was identical in the two treatment arms and did not influence the SUA-AF relationship.

In parallel Cox models that were adjusted for the same covariates, new-onset AF defined by physician reports (n = 531) was independently predicted by time-varying SUA level [HR = 1.10 per 16.8 μmol/L (1 mg/dL), (95% CIs, 1.04–1.16)], as was the first occurrence of AF defined by either Minnesota code or physician reports (n = 644) [HR = 1.11 per 16.8 μmol/L (1 mg/dL), (95% CIs, 1.06–1.16), both P < 0.0001].

In a parallel Cox model that adjusted for the same covariates, new-onset AF defined by physician reports (n = 535) was higher in the 4th quartile of time-varying SUA level [HR = 1.40 (95% CIs, 1.07–1.83), P = 0.015] independent of effects of history of cerebrovascular disease [HR = 1.30 (95% CIs, 1.08–1.56), P = 0.006], age [HR = 1.65 per 7.0 years (95% CIs, 1.50–1.82), P ≤ 0.0001], and male sex [HR = 1.51 (95% CIs, 1.19–1.91), P = 0.001]. Higher systolic BP, Cornell voltage-duration product, Framingham risk score, and treatment with losartan did not contribute significantly to this multivariable analysis. In another Cox model that adjusted for the same covariates (Figure 1), new-onset AF defined by either Minnesota code or physician reports (n = 652) was higher in the 4th quartile of time-varying SUA level [HR = 1.40 (95% CIs, 1.10–1.79), P = 0.007] independently of effects of greater age [HR = 1.67 per 7.0 years (95% CIs, 1.53–1.83), P ≤ 0.0001], male sex [HR = 1.45 (95% CIs, 1.17–1.80), P = 0.001], and history of cerebrovascular disease [HR = 1.32 (95% CIs, 1.11–1.56), P = 0.001]. The Cornell voltage-duration product, higher systolic BP, Framingham risk score, and losartan treatment did not contribute significantly to this multivariable analysis.